Skip to content

getting-started.md

Latest commit

 

History

History
67 lines (47 loc) · 1.8 KB

getting-started.md

File metadata and controls

67 lines (47 loc) · 1.8 KB

Getting Started

Install

git clone https://github.com/txmed82/case-crawler.git
cd case-crawler
pip install -e ".[dev]"

Optional extras:

pip install -e ".[hf,imaging,parquet]"

Use hf for Hugging Face dataset/model helpers, imaging for image generation and image-text validators, and parquet for parquet export.

Generate A Dataset Without API Keys

casecrawler generate-dataset "sepsis" --count 10
casecrawler datasets
casecrawler datasets quality <dataset_id>

The default path is deterministic and offline. It creates structured EHR facts, labs, vitals, clinical documents, validation reports, and export-ready records without requiring an LLM key.

Generate A Multimodal Release Package

casecrawler generate-release-package "mixed acute care cohort" \
  --count 25 \
  --max-validation-retries 2 \
  --output-dir release-package \
  --format multimodal_jsonl \
  --seed casecrawler

casecrawler verify-split-package --require-multimodal-release release-package

generate-release-package runs the full multimodal recipe, seeds bundled reference fixtures, runs benchmark gates, writes train/validation/test splits, copies file-backed image and time-series artifacts, creates dataset/model cards, and verifies the package.

Run The API And UI

casecrawler serve

Or with Docker:

cp .env.example .env
docker compose up

The API defaults to http://localhost:8000. The frontend is served by the Docker stack.

Human Review

Datasets generated with --require-human-review are blocked from export until records are approved:

casecrawler generate-dataset "sepsis" --count 5 --require-human-review
casecrawler reviews queue --dataset-id <dataset_id>
casecrawler reviews mark <record_id> --status approved --reviewer clinical-reviewer